Intelligent Automation for AI-driven Document Understanding

Abstract: While machine learning models are continually improving, for most tasks they fail to achieve perfect predictive performance. In order to be a valuable tool in decision-making under uncertainty, it stands to reason that we want some statistical guarantees on the quality of probabilistic predictive models. Research into calibration regained popularity after repeated empirical observations … Read more

Multimedia content protection: from model-based to data-driven approaches

Abstract: In this talk, Dr Andrew D. Bagdanov, Dr Massimo Iuliani, Dr Dasara Shullani, Dr Daniele Baracchi & Simone Magistri will present the results of the research activity on multimedia content security carried out at the Information Engineering Department (DINFO) at the University of Florence. First, we will give a short introduction to the three … Read more

Interpretable-by-design Prototype-based Deep Learning

Abstract: Deep Learning justifiably attracted the attention and interest of the scientific community and industry as well as of the wider society and even policy makers. However, the predominant architectures (from Convolutional Neural Networks to Transformers) are hyper-parametric models with weights/parameters being detached from the physical meaning of the object of modelling. They are, essentially, … Read more

Meta-RL for Visual Semantic Navigation

Abstract: Semantic and goal-oriented visual navigation is one of the most prominent tasks performed by intelligent species in their daily lives. This task is defined as the ability we have to navigate through our environment, finding targets and enabling interaction with it. Navigation methods used in robotics can be divided into two main categories: geometry-based … Read more

Self-supervised Learning from Images, and Augmentations

Abstract:  In this talk, Yuki M. Asano will talk about pushing the limits of what can be learnt without using any human annotations. After a first overview of what self-supervised learning is, we will first dive into how clustering can be combined with representation learning using optimal transport ([1] @ ICLR’20), a paradigm still relevant … Read more

Mining large medical radiology image repositories

Abstract: Developing clinical predictive models by processing medical radiology images is often challenging due to high variability of data, noise and data scarcity. Using pre-trained feature extractors in deep learning configurations to initialise weights is often beneficial to the model optimisation process, leading to faster convergence and more accurate models. Although one can also benefit … Read more

Improving generalization for classification and retrieval tasks

Abstract: In this talk we will present recent works that generally aim at improving the generalization of visual representations on both classification and retrieval tasks. We will start from a recent work on supervised pre-training and will present an approach that aims to improve the transferability of encoders learned in a supervised manner, while retaining … Read more

Deep Learning for Document, Scene and Satellite Images Processing and Recognition

Abstract: Deep learning applications have been thriving over the last decade in many different domains, including image processing and recognition. The driver for the vibrant development of deep learning have been the availability of abundant data. This talk reviews the main results of our research activities carried out over the last few years. During this … Read more

From scans to information: end-to-end information extraction from documents

Abstract: With the advancement of Transformers, especially as far as computer vision is concerned, we are starting to apply end-to-end neural networks, without OCR or other pre-/postprocessing techniques, to the challenges of document understanding and information extraction. I will present developments in this area and discuss potential problems from both theoretical (handling longer sequences) and … Read more

Calibrated Fine-Grained Recognition and Retrieval

Abstract: In the last decade, many areas of computer vision have progressed to a level supporting reliable, and sometimes impressive, applications. I wil talk about two such domains, fine-grained recognition and visual retrieval. In the fine-grained recognition, I’ll discuss the issue of prior probability shift, classifier calibration and the choice of loss functions driven by … Read more